def mds_loader(params, is_key_frame=True):
# Datasets
# TODO: put the path to your train, test, validation txt files
if is_key_frame:
label_file_train = 'dataloader_files/keyframe_data_train.txt'
label_file_val = 'dataloader_files/keyframe_data_val.txt'
else:
label_file_train = 'dataloader_files/videoframe_data_train.txt'
label_file_val = 'dataloader_files/videoframe_data_val.txt'
label_file_test = 'dataloader_files/videoframe_data_test.txt'
mean = [0.5, 0.5, 0.5]
std = [0.5, 0.5, 0.5]
train_dataset = Mds189(label_file_train,loader=default_loader,transform=transforms.Compose([
transforms.ColorJitter(hue=.05, saturation=.05),
transforms.RandomHorizontalFlip(p=0.33),
transforms.RandomRotation(degrees=15),
transforms.ToTensor(),
transforms.Normalize(mean, std)
]))
train_loader = data.DataLoader(train_dataset, **params)
val_dataset = Mds189(label_file_val,loader=default_loader,transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean, std)
]))
val_loader = data.DataLoader(val_dataset, **params)
if is_key_frame:
return train_loader, val_loader
elif not is_key_frame:
test_dataset = Mds189(label_file_test,loader=default_loader,transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean, std)
]))
test_loader = data.DataLoader(test_dataset, **params)
return train_loader, val_loader, test_loader
label_file_train = 'dataloader_files/videoframe_data_train.txt'
label_file_val = 'dataloader_files/videoframe_data_val.txt'
label_file_test = 'dataloader_files/videoframe_data_test.txt'
####################### NORMALIZE + DATA GEN #######################
if is_key_frame:
mean = np.array([134.010302198, 118.599587912, 102.038804945]) / 255
std = np.array([23.5033438916, 23.8827343458, 24.5498666589]) / 255
else:
mean = np.array([133.714058398, 118.396875912, 102.262895484]) / 255
std = np.array([23.2021839891, 23.7064439547, 24.3690056102]) / 255
# Generators
train_dataset = Mds189(label_file_train,
loader=default_loader,
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean, std)
]))
train_loader = data.DataLoader(train_dataset, **params)
val_dataset = Mds189(label_file_val,
loader=default_loader,
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean, std)
]))
val_loader = data.DataLoader(val_dataset, **params)
####################### TRAIN/VAL #######################
start = time.time()
print('Beginning training..')
]
std = [23.2021839891 / 255.0, 23.7064439547 / 255.0, 24.3690056102 / 255.0]
# TODO: you should normalize based on the average image in the training set. This shows
# an example of doing normalization
# TODO: if you want to pad or resize your images, you can put the parameters for that below.
# Generators
# NOTE: if you don't want to pad or resize your images, you should delete the Pad and Resize
# transforms from all three _dataset definitions.
train_dataset = Mds189(
label_file_train,
loader=default_loader,
transform=transforms.Compose([
# transforms.Pad(requires_parameters), # TODO: if you want to pad your images
# transforms.Resize(requires_parameters), # TODO: if you want to resize your images
transforms.RandomAffine((-30, 30), shear=(-10, 10), fillcolor=0),
transforms.ToTensor(),
transforms.Normalize(mean, std)
]))
train_loader = data.DataLoader(train_dataset, **params)
val_dataset = Mds189(
label_file_val,
loader=default_loader,
transform=transforms.Compose([
# transforms.Pad(),
# transforms.Resize(),
transforms.ToTensor(),
transforms.Normalize(mean, std)
]))
label_file_train = '../data/hw6_mds189/videoframe_data_train.txt'
label_file_val = '../data/hw6_mds189/videoframe_data_val.txt'
label_file_test = '../data/hw6_mds189/videoframe_data_test.txt'
# TODO: you should normalize based on the average image in the training set. This shows
# an example of doing normalization
mean = [0.5, 0.5, 0.5]
std = [0.5, 0.5, 0.5]
# TODO: if you want to pad or resize your images, you can put the parameters for that below.
# Generators
# NOTE: if you don't want to pad or resize your images, you should delete the Pad and Resize
# transforms from all three _dataset definitions.
train_dataset = Mds189(label_file_train,loader=default_loader,transform=transforms.Compose([
transforms.Pad(requires_parameters), # TODO: if you want to pad your images
transforms.Resize(requires_parameters), # TODO: if you want to resize your images
transforms.ToTensor(),
transforms.Normalize(mean, std)
]))
train_loader = data.DataLoader(train_dataset, **params)
val_dataset = Mds189(label_file_val,loader=default_loader,transform=transforms.Compose([
transforms.Pad(),
transforms.Resize(),
transforms.ToTensor(),
transforms.Normalize(mean, std)
]))
val_loader = data.DataLoader(val_dataset, **params)
if not is_key_frame:
test_dataset = Mds189(label_file_test,loader=default_loader,transform=transforms.Compose([
transforms.Pad(),